From the presentation of the 1st report of #Mexico's six year term headed by #Presidenta Claudia #Sheinbaum:
"The dark night of #neoliberalism, a model that established that the state should not intervene in development nor concern itself with the redistribution of wealth, but simply create a favourable environment for business, is now a thing of the past; the experience of those decades demonstrated that this idea was totally erroneous."
(my translation)
@jimkreft
If there is, I have not seen it. There's of course the transcript of the presentation on a certain video streaming platform and that thing has a auto translation function. It has been possible in the past to obtain those transcripts as text. 😉
@keith yeah, but that’s all LLM-driven garbage, especially for a text like this that is unlike most of the text on the internet. The bias inherent in an LLM will not result in a good translation of this report.
@jimkreft
Indeed, machine translation* might miss some subtleties
However, even as I eschew big tech such as alphabet inc (for everything they've done to the internet), I'd still prefer that somebody with no knowledge at all of the Spanish language would watch the presentation with translated subtitles, than (only) read what is out there in the English language from "journalists" on the subject. Megan Janetsky for example. 😔
Anyway, it seems eng MT subs are disabled on it.
* see next toot.
@jimkreft
You know, for years now, since initial attempts that were often laughable, I've been OK with machine translation, generally finding it to be faithful enough in languages I know and useful when I would otherwise be at a complete loss; your post prompted me to look into #LLM vs MT and I see that things are moving in the direction of implementing translation more using LLM tech. I haven't actually done any analysis of LLM translation they way I did before with MT when I came to trust it.
@keith it’s quite different, and fails in ways that are much harder to recognize. Machine transcription and translation (boring old machine learning, natural language processing, and bounded domain relevant models)is indeed quite good, and also relatively easy to recognize when it messes up. On the other hand, LLM driven transcription and translation generates output that LOOKS fluent, but often makes up entire passages that are driven by the model rather than by the actual source content.
@keith I also cannot emphasize the bias enough. The bias part is also extremely important. LLMs are based on the whole internet (plus most books), which on the whole has a western, capitalist, patriarchal bias. There is no way I would trust such an unbounded model to transcribe and translate an anti neoliberal text from the global south. I recommend following @emilymbender and @timnitGebru among others to get an expert viewpoint on these kinds of things.
@jimkreft
Indeed, pasting something into deepl and comparing the difference between Classic Language Model (pretty good translation) and AI "next-gen" language model - the latter is quite inaccurate.
@keith do you know if there is an English translation of the whole thing kicking around somewhere?